AI-based decision support tool can reduce unnecessary breast biopsies, save millions

A deep learning-based clinical decision support tool may help reduce unnecessary breast biopsies, according to a new study published Aug. 9 in Radiology: Artificial Intelligence.

Scientists at Houston Methodist Hospital previously developed the tool, called iBRISK for intelligent-augmented breast cancer risk calculator, to better gauge women’s danger of developing the disease. Scientists created the tool by applying deep learning to clinical risk factors and mammographic descriptors from 9,700 individuals imaged at their institution, validating it on another 1,000-plus patients.

In a new investigation, scientists further tested iBRISK on an independent, retrospective dataset of breast images from 4,209 more patients, screened across three Texas institutions between 2006 and 2016. The model was specifically developed to assess the probability of malignancy of Breast Imaging Reporting and Data System (BI-RADS) category 4 lesions

Lead author Chika F. Ezeana, with the Houston Methodist Neal Cancer Center, and colleagues found promising results, estimating the AI tool could save many millions of dollars in avoided exams.

“The iBRISK calculator can assist physicians, primarily radiologists, in triaging patients to low-probability of malignancy (POM) groups to avoid biopsies of benign lesions, while high-risk groups can be treated as BI-RADS 5 patients,” the authors reported. “A more precise stratification system that considers vital patient characteristics in addition to abnormal, suspicious imaging features is needed to enhance POM estimation to guide the safe management of such mammographic findings, prevent over-biopsy and associated costs, and reduce patient emotional distress.”

The iBRISK model’s accuracy was calculated at about 89.5%, with an area under the receiver operating characteristic curve of 0.93. Specificity was 81%, and only two of the 1,228 individuals in the “low” POM group had malignant lesions. Conversely, the “high” POM malignancy rate was 85.9%. IBRISK score as a continuous predictor of malignancy yielded an area under the receiver operating characteristic curve of 0.97.

Ezeana et al. estimated iBRISK could potentially save hundreds of millions each year at a single institution. This is based on a Medicare biopsy reimbursement rate of $380 and the average cost for each different type of such tests. The most common type of biopsy at MD Anderson Cancer Center is stereotactic (68%), while the cost for different types can range from $321 to nearly $3,600 for a mammography-guided surgical exam. They believe cost savings of some $420 million could be reached across a sample of 390,000 women due to biopsy avoidance.

“Our study demonstrates that iBRISK can effectively aid in risk stratification of BI-RADS 4 lesions and reduce overbiopsy of these lesions,” the authors concluded. “Ultimately, the iBRISK calculator will be published as an online interface and made open access, noncommercial, and accessible by health systems and centers worldwide. Future studies aim to improve the model further, particularly by including more granular data and other BI-RADS categories.”